Message Boards

GROUPS:

Shenghui Yang

Work with Julia Image Data in WL

Shenghui Yang, WOLFRAM

Posted 15 days ago

162 Views

0 Replies

2 Total Likes

Follow this post

Motivation

Some tips for image conversion and data serialization.
For anyone interested in image processing in both WL and Julia for Introduction to Computational Thinking 18.S191 Fall 2020 course. Youtube Link here.

Prerequist

I install Julia-1.5.3 via homebrew on macOS for instance

$ brew install julia

Once installation is done I follow this instruction to install ZMQ and JSON in a terminal with Julia running. After they are installed, I can run RegisterExternalEvaluator to link Julia to WL

In[]:=

RegisterExternalEvaluator["Julia","/usr/local/bin/julia"]

Out[]=

7d0829b2-a95d-4780-8f7f-92e38dced69b

Verify registration

In[]:=

FindExternalEvaluators["Julia"]

Out[]=

	System	Version	Target	+	Registered
7d0829b2-a95d-4780-8f7f-92e38dced69b	Julia	1.5.3	/usr/local/bin/julia	•	True
25eefddc-ad9b-4fe4-8a81-acbdba98837d	Julia	1.5.3	/Applications/Julia-1.5.app/Contents/Resources/julia/bin/julia	•	Automatic

Check the version

〉

In[]:=

VERSION

Out[]=

major1,minor5,patch3,prerelease{},build{}

Download extra packages

〉

Pkg.add("Images")
Pkg.add("QuartzImageIO")

Add the proper name space and modules before using Images package

〉

In[]:=

using Images

Check the current linked Julia session running on my machine

In[]:=

juliaSession=ExternalSessions[][[1]]

Out[]=

ExternalSessionObject

System: Julia	Version: 1.5.3
Name: DefaultJuliaSession



Import Image and Change Orientation

Sample input file from MIT computational thinking course 1.

〉

In[]:=

url = "https://i.imgur.com/VGPeJ6s.jpg"
download(url,"philip.jpg")

Out[]=

philip.jpg

You can either import using Julia or WL. When I import image from Julia, I convert the image object to raw double array first with Julia’s ImageCore.channelview function. If no conversion is specified, ZeroMQ will convert image pixel data into a very large list of millions lines of dictionary/association for JSON file, then WL imports the JSON file to display it with some format rearrangement. This data serialization-and-deserialization process take rather long time

In[]:=

imgDataJulia=ExternalEvaluate[juliaSession,"img = load(\"philip.jpg\");channelview(img)"];

I can save some notebook file size and improve response by rasterizing output image automatically with $Post. This only affect output. values in variables in imgJulia1 is intact

In[]:=

$Post=If[MatchQ[#,_Image|_ByteArray|_NumericArray],Rasterize[#,RasterSize300],#]&

Out[]=

If[MatchQ[#1,_Image|_ByteArray|_NumericArray],Rasterize[#1,RasterSize300],#1]&

In[]:=

imgJulia1=Image[imgDataJulia,ColorSpace"RGB"]

Out[]=

Notice that the convention is slightly different between WL and Julia. Comparing to what’s shown in the online video or direct import with WL code , Corgi’s head point downward instead of right.

In[]:=

imgWL=Import["https://i.imgur.com/VGPeJ6s.jpg"]

Out[]=

I need to swap the first and second dimension of image data from Julia’s color channel multi-dimensional array

In[]:=

ImageData[imgWL]//Dimensions

Out[]=

{3675,2988,3}

In[]:=

ImageData[imgJulia]//Dimensions

Out[]=

{2988,3675,3}

Now this is the image with correct orientation after transposition

In[]:=

imgJulia2=Image[Transpose[imgDataJulia,{2,1,3}],ColorSpace"RGB"]

Out[]=

Save and Share Large Array

High definition image is a large numeric array. I found the following function really useful in keep such data persistently on local machine and sharing across internet for data integrity purpose. WL automatically translate ImageData to packed array

In[]:=

Developer`PackedArrayQ[ImageData[imgWL]]

Out[]=

True

I can use NumericArray as well which also support functions like Take, Part or ListPlot

In[]:=

NumericArray[ImageData[imgWL],"Real32"]

Out[]=

If I would like to save the data on disk or within notebook in a compact manner, either of function below works and I can export a WXF binary file later

In[]:=

bin=BinarySerialize[ImageData[imgWL],PerformanceGoal"Size",Method{"PackedArrayRealType""Real32"}]

Out[]=

In[]:=

ExportByteArray[ImageData[imgWL],"WXF",PerformanceGoal"Size",Method{"PackedArrayRealType""Real32"}]

Out[]=

To save the ByteArray object, I just need to write them into a file

In[]:=

myfile=CreateFile[FileNameJoin[{$HomeDirectory,"test1"}]];BinaryWrite[myfile,bin];Close[myfile]

Out[]=

/Users/shenghuiy/test1

Import it again to check if the file is saved properly

In[]:=

Import[myfile]//Image

Out[]=

Restore $Post to no definition so the front end no longer rasterize image output automatically

In[]:=

$Post=.

Summary

◼

Install Julia from your favorite repository and package management system

◼

Transposition of Image data is needed between image process functions in both languages

POSTED BY: Shenghui Yang

Answer

Reply to this discussion

Reply Preview

Attachments

Remove Add a file to this post

Follow this discussion

or Discard

Group Abstract

Feedback

Enable JavaScript to interact with content and submit forms on Wolfram websites. Learn how »